Amin Beheshti
Professor Amin Beheshti is a Full Professor of Data Science at Macquarie University in Sydney, Australia. He is the founder and director of the Centre for Applied Artificial Intelligence, head of the Data Science Lab, and founder of the Big Data Society at Macquarie University. He is also the founder and CEO of Collective Learning AI (https://collectivelearning.ai/). Amin received Prestigious Awards, including the AI Academic/Researcher of the Year at the inaugural Australian National AI Awards 2024. Alongside his teaching and leadership roles, he has made significant contributions to applied AI and data science research, securing more than 70 research projects and over $50 million in research funding. He has supervised more than 50 research students, with his PhD graduates now leading AI initiatives at global technology companies including Microsoft and Google. His long-standing vision is to shift AI from a technology of endless theoretical potential to one that is actively, intentionally, and responsibly applied to solve meaningful real-world problems.
Mored details can be found here: https://data-science-group.github.io/people/aminbeheshti/
Title: Neural Processes: Toward a Knowledge-First Foundation for BPM in the Age of Generative and Agentic AI
Abstract: Business Process Management has traditionally been shaped by a model-first assumption: before a process can be enacted, governed, analysed, or improved, it should first be represented as an explicit process model. This keynote introduces “Neural Processes”, a knowledge-first foundation for BPM in the age of generative and agentic AI. Inspired by Neural Databases, which challenged the need for predefined data schemas by enabling reasoning over natural language and neural representations, Neural Processes ask a parallel question for BPM: what if process behaviour were not primarily derived from predefined control-flow models, but generated, constrained, justified, and audited through reasoning over organizational knowledge, historical executions, contextual goals, policies, and domain expertise?
The keynote develops Neural Processes around three pillars: (i) a language-native Process Knowledge Graph (KG), referred to as “Liquid KG”, that captures organizational expertise, execution histories, regulations, exceptions, decisions, and best practices as semantically linked knowledge assets; (ii) a neural process reasoning engine, referred to as “ProcessGPT”, that uses large language models, retrieval, graph reasoning, tool use, and policy constraints to infer and adapt process behaviour at runtime; and (iii) a governed execution layer, formalized as “Agentic Service-Oriented Computing”, in which process instances are realized as orchestrated ensembles of AI agents engineered as composable, observable, secure, auditable, and policy-compliant services. Neural Processes are not simply a rebranding of LLM-based workflow automation: they propose a BPM foundation in which process knowledge, rather than the process model alone, becomes the primary artifact. The keynote grounds this vision in a knowledge-intensive enterprise scenario and concludes by mapping the open research challenges around correctness, conformance, explainability, provenance, human oversight, agent governance, and the future relationship between BPM, service computing, and generative AI.